Classification And Dynamic Analysis Of Typical Pole-like Objects In Road Scenes Based On Vehicle-borne LiDAR Data

Typical ground objects in road scenes include street lamps,street trees,traffic signs,telegraph poles and other pole-like objects.As an important part of urban road scene,the rapid and automatic extraction and classification of pole-like objects is of great significance to the construction of digital and smart city.Traditional surveying and mapping methods not only need to spend a lot of manpower and financial resources,but also have very low efficiency.As a high-efficiency 3D geospatial information acquisition technology,the vehicle-borne light detection and ranging system(vehicle-borne LiDAR)can quickly obtain detailed 3D spatial information of road and ground objects on both sides,and the spatial characteristics of pole-like objects can be accurately expressed in the three-dimensional point cloud,providing a new technical support for obtaining the pole-like object information.The extraction and classification of typical ground objects in vehicle-borne LiDAR point cloud data has been a research hotspot for scholars at home and abroad.Therefore,through analyzing the research status at home and abroad,the typical ground objects in urban road scenes were taken as research objects,and a series of related researches were carried out around the extraction,classification,modeling and dynamic change of ground objects.The specific research work is as follows:(1)The working principle and development history of vehicle-borne LiDAR were briefly introduced.Based on the analysis of the research status at home and abroad,the deficiencies in the research on the extraction and classification of typical features in urban road scenes were summarized.In view of the deficiencies of current research,the research ideas and research routes were determined.(2)Experimental data and preprocessing.The experimental area was determined and the data was collected.The preprocessing of data clipping and partitioning was carried out as required.The improved algorithm based on mathematical morphology was used to extract the ground points and the rule DEM was constructed according to the surface points.And the DEM model was used to construct elevation standard to filter low ground objects.(3)Extraction and classification of typical ground objects.The remaining point clouds were clustered and segmented,and the ground features were divided into independent units.Typical pole-like objects(street lamps,street trees,traffic signs and telegraph poles)were extracted according to the spatial characteristics of pole-like objects and other methods to prepare for the classification of ground objects.According to the morphological characteristics of different ground objects,several eigenvalues that could effectively distinguish pole-like objects were selected.And then the eigenvalues of ground object clustering unit were obtained and the eigenvectors were formed.The BP neural network classification model was used to classify the ground objects.(4)Dynamic analysis of typical ground objects.The relative accuracy of typical ground objects in vehicle-borne LiDAR point cloud data was verified.Multi-period data in the same region was collected and dynamically analyzed.According to the specific information of the location,height and inclination of the ground objects,the changes of ground objects were analyzed.(5)Verify the feasibility of the algorithm.The vehicle-borne LiDAR point cloud data in the experimental area was selected to verify the feasibility of the algorithm through experiments such as data preprocessing,typical ground object extraction,typical ground object classification and dynamic analysis of multi-period data.